[Bug middle-end/50731] [4.7 Regression] FAIL: gcc.dg/torture/vector-shift2.c

rguenth at gcc dot gnu.org gcc-bugzilla@gcc.gnu.org
Thu Oct 27 11:01:00 GMT 2011


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50731

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1
             Status|NEW                         |ASSIGNED
         AssignedTo|unassigned at gcc dot       |rguenth at gcc dot gnu.org
                   |gnu.org                     |

--- Comment #5 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-10-27 11:00:49 UTC ---
I suppose it's just completely optimized away at -O1+ and thus broken
unconditionally.  Shift lowering gives us

  vuchar_1 = { 1, 2, 3, 4, 1, 2, 3, 4, 1, 2, 3, 4, 1, 2, 3, 4 };
  vchar0_2 = { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 };
  vchar1_3 = { -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
};
  D.2720_7 = BIT_FIELD_REF <vchar1_3, 8, 0>;
  D.2721_8 = D.2720_7 << 1;
  D.2722_9 = BIT_FIELD_REF <vchar1_3, 8, 8>;
  D.2723_10 = 1;
  D.2724_11 = BIT_FIELD_REF <D.2723_10, 8, 8>;

and that last two lines already look completely bogus...

We are lowering from

  vuchar_1 = { 1, 2, 3, 4, 1, 2, 3, 4, 1, 2, 3, 4, 1, 2, 3, 4 };
  vchar0_2 = { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 };
  vchar1_3 = { -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
};
  i1.0_4 = vchar1_3 << 1;
  i1 = i1.0_4;
  D.2714_5 = BIT_FIELD_REF <i1, 8, 8>;

but the generic binary lowering and do_binop do not handle a scalar
RHS separately but do

static tree
do_binop (gimple_stmt_iterator *gsi, tree inner_type, tree a, tree b,
          tree bitpos, tree bitsize, enum tree_code code)
{
  a = tree_vec_extract (gsi, inner_type, a, bitsize, bitpos);
  b = tree_vec_extract (gsi, inner_type, b, bitsize, bitpos);
  return gimplify_build2 (gsi, code, inner_type, a, b);

that of course only works for element 0.

We came here via

  gcc_assert (code != VEC_LSHIFT_EXPR && code != VEC_RSHIFT_EXPR);
  new_rhs = expand_vector_operation (gsi, type, compute_type, stmt, code);

(gdb) p code
$6 = LSHIFT_EXPR

Fixing it in do_binop like

static tree
do_binop (gimple_stmt_iterator *gsi, tree inner_type, tree a, tree b,
          tree bitpos, tree bitsize, enum tree_code code)
{
  if (TREE_CODE (TREE_TYPE (a)) == VECTOR_TYPE)
    a = tree_vec_extract (gsi, inner_type, a, bitsize, bitpos);
  if (TREE_CODE (TREE_TYPE (b)) == VECTOR_TYPE)
    b = tree_vec_extract (gsi, inner_type, b, bitsize, bitpos);
  return gimplify_build2 (gsi, code, inner_type, a, b);

is easiest.



More information about the Gcc-bugs mailing list